Dense surface reconstruction using a learning-based monocular vSLAM model for laparoscopic surgery.

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Augmented reality (AR) is increasingly used in surgery for overlaying pre-operative imaging onto the surgical field.
• Accurate registration of deformable anatomy, especially with monocular laparoscopes in minimally invasive surgery, remains a challenge.
• Robust surgical scene reconstruction is crucial for precise AR-guided procedures and related applications.

Purpose of the Study:

• To develop and evaluate a deep-learning-based visual simultaneous localization and mapping (vSLAM) algorithm for 3D surgical scene reconstruction.
• To enable accurate camera pose estimation and dense 3D reconstruction from monocular laparoscopic video data.
• To establish a framework for evaluating vSLAM algorithms on challenging surgical surfaces.

Main Methods:

• Utilized a state-of-the-art deep-learning vSLAM algorithm to process monocular laparoscope video.
• Generated dense 3D reconstructions, camera pose estimations, and depth maps in real-time.
• Developed a framework for evaluating vSLAM performance on non-Lambertian, low-texture surgical surfaces.

Main Results:

• The vSLAM method successfully generated robust 3D reconstructions and accurate camera pose estimations from monocular laparoscopic video.
• The system provided these crucial outputs without requiring stereo vision or additional sensors, enhancing usability.
• Demonstrated the utility of the evaluation framework for assessing vSLAM algorithms in surgical contexts.

Conclusions:

• The proposed deep-learning vSLAM approach offers a viable solution for accurate 3D surgical scene reconstruction using monocular laparoscopes.
• This method improves the foundation for advanced AR-guided surgery, remote assistance, and surgical simulation.
• The developed evaluation methods will aid in refining future vSLAM algorithms for surgical applications.